KR20240063385A - Skin marker position detection system - Google Patents

Abstract

Treatment planning systems for electrical stimulation therapy often determine the position and angle of electrode pads relative to a three-dimensional model. However, finding the position and angle on the skin surface of an actual subject and attaching the electrode pad is not an easy task, and therefore, assistance from an electrode guide system that provides position and angle information of the electrode pad is required. The core of the electrode guide system is the provision of a reference coordinate system that equally corresponds to the object and the three-dimensional model. In the present invention, a skin marker or stereotactic device is attached to the surface of an object, the measured medical image is input into a pre-trained machine learning system, and the coordinates of the center point of the skin marker or stereotactic device are output to the surface of a three-dimensional model. A method for providing a reference coordinate system is published.
A method for identifying skin markers of an object using AI is provided. The skin markers include natural skin markers that can be found inside the human body and artificial skin markers made from artificial materials.

Description

Skin marker coordinate detection system {Skin marker position detection system}

The present invention relates to a method and system for moving a center point and detecting coordinates on the surface of a tissue type based on the center point of a skin marker for electrical stimulation therapy, using a skin marker coordinate detection system.

Electrical stimulation therapies such as transcranial direct current stimulation (tDCS), transcranial magnetic stimulation (TMS), and tumor-treating fields (TTFields) are non-invasive treatment methods. , It is widely used to treat various diseases depending on the location and intensity of stimulation. tDCS, TMS, and TTFields treatments can create an optimal treatment plan tailored to the patient's individual characteristics through treatment planning system software. However, there is no means to accurately map the position and angle of the electrode array in the treatment plan determined in the treatment planning system to the position and angle to be attached to the actual patient, making it difficult for doctors to perform the procedure by attaching the electrode array to the correct position and angle. There is. Therefore, in order to increase treatment effectiveness, increase safety, and reduce risk, a method of attaching the electrode array as intended is needed while simultaneously considering location and angle. Here, the electrode array is attached using a medical skin marker. Although the skin marker can be identified in the medical image, the location where it is actually attached to the patient is different because coordinates based on the skin marker are used. Therefore, a system that detects the center point based on the skin marker as the center point on the patient's body is needed.

The problem to be solved by the present invention is to provide a skin marker coordinate detection system.

The marker position detection system includes the following steps: marking a first reference point on an object; attaching a skin marker based on the reference point; acquiring a medical image including a skin marker; inputting a machine learning medical image; learning machine learning. A data generation step including; A step of acquiring skin marker and biological tissue segmentation data; A step of acquiring a center point of a skin marker based on the segmentation data; A step of detecting a second reference point based on the obtained center point; A data acquisition step including; It includes a data output step including a step of outputting skin marker and biological tissue segmentation data using learned machine learning; and a step of outputting second reference point information of the skin marker.

In one embodiment of the present invention, the indication of the first reference point includes a temporary tattoo, tattoo, sticker, or skin marker.

In one embodiment of the present invention, in the method of displaying the center point of a skin marker, the number of center points displayed on the object is one or more, and in the method of displaying the center point of the skin marker, the number of center points displayed as the center point that can be used is It may include the center point when the first reference point is viewed vertically and the center of gravity of the skin marker.

In one embodiment of the present invention, a method for identifying the center point in medical images (CT, MRI, CT, PET, SPECT, FMRI (Functional MRI)) includes a method of identifying the location based on the size, shape, and intensity of the marker. can do.

In one embodiment of the present invention, in the method of acquiring a medical image, the medical image may include MRI, CT, PET, SPECT, and FMRI (Functional MRI).

According to the present invention, medical images including skin markers are acquired, a model of the tissue type and skin marker is created using machine learning, and based on this, the coordinates of the center point are detected and the coordinates of the segmented tissue type surface are accurately located. Electrodes can be attached, and the electrode attachment location of the treatment plan can be matched with the electrode attachment location of the target.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below.

1 is a flowchart explaining a skin marker coordinate detection system according to this embodiment.
Figure 2 is a diagram showing a problem with current technology.
Figure 3 shows that in detecting the second reference point for the purpose of this study, the center point of the skin marker cannot be found because the existing machine learning technology classifies the skin marker by tissue type (skin).
Figure 4 is a diagram explaining that the stage of the present invention consists of (1) a segmentation stage and (2) a reference point detection stage (Fine Stage).
Figure 5 is a diagram illustrating the process (projection) of moving a point of the center point of a skin marker to a point on the skin.
Figure 6 shows the skin marker, the center point of the skin marker (The Feature Point of Skin Marker), and the detected second reference point (The Second Reference Point on the Skin) detected in the medical image using the method presented in the present invention. It's a picture.
Figure 7 is a diagram showing an example in which the detected second reference point is used.
Figure 8 is a diagram showing the data output step to be output to the user.
Figure 9 shows two methods for applying machine learning to detect the second reference point.

1 is a flowchart explaining a skin marker coordinate detection system according to this embodiment.

This method is largely divided into a data input step, a data generation step, a data acquisition step, and a data output step. The data input step is a skin marker attachment step based on a reference point, a skin marker attachment step based on a reference point, and a medical image acquisition step including a skin marker. It is divided into, and the first reference point marking step is a step of marking reference points used in the treatment plan on the subject's body, and one or more may be displayed. A step of attaching a skin marker based on a reference point attaches a skin marker on the marked reference point. Since skin markers can be identified in medical images, they are used to obtain feature points to identify the first reference point. In the image acquisition step including a skin marker, a medical image captured with a skin marker attached to the medical image is acquired.

The data generation step consists of a machine learning medical image input step and a machine learning learning step. The machine learning learning step is a step to create a machine learning model that generates data divided by tissue type using medical images. The generated machine learning model is passed to the machine learning medical image input stage, the model created from the machine learning learning stage is stored, and when new data comes in, the stored model is used to provide results according to learning.

The data acquisition step consists of a skin marker and biological tissue segmentation data acquisition step, a skin marker center point acquisition step based on the segmentation data, and a second reference point detection step based on the obtained center point. In the skin marker and biological tissue segmentation data acquisition step, learned machine learning is used to input a medical image containing the obtained skin marker and obtain biological tissue segmentation data. The acquired segmented data is divided into biological tissues (skin, skull, cerebrospinal fluid, gray matter, white matter, etc.) and skin markers. Additionally, this segmented tissue segmentation data can be used to create a three-dimensional volume area of interest (skin marker, biological tissue). The step of acquiring the center point of the skin marker based on the segmentation data is a step of detecting the center point using the obtained segmentation data. Here, the center point means a point that can represent the skin marker. Specifically, it may include the center of gravity and the center point of the bounding box. This center point can be detected using mathematical formulas or volume data. The second reference point detection step based on the obtained center point is a step of detecting the second reference point on the data divided into tissue types based on the acquired skin marker center point. In this embodiment, projection is used to detect the second reference point. A second reference point based on the marker was detected. The coordinates of the second reference point detected in this way can be used in the treatment planning system, and the position coordinates of the electrode or electrode array can be generated based on the second reference point.

The data output step consists of a skin marker and biological tissue segmentation data output step using learned machine learning, and a second reference point information output step of the skin marker. The skin marker and biological tissue segmentation data output step using learned machine learning is described above. This is the step of outputting the skin markers of the data acquisition step and the segmentation data obtained in the biological tissue segmentation data acquisition step. The output can be output in the form of a 3D Volume Mesh or in a flat form such as Axial, Coronal, or Sagitta. The step of outputting the second reference point information of the skin marker is the step of outputting the obtained second reference point, and information on the detected second reference point. By outputting to the user, the coordinates of the second reference point can be used in the treatment planning system, and the position coordinates of the electrode or electrode array can be generated based on the second reference point.

Figure 2 is a picture showing the problem of the current technology. In general, the method in which skin markers are used is to mark the body of the object (permanent tattoos, temporary tattoos, stickers, etc.) and attach a skin marker thereon. Next, the medical image including the skin marker is input, the marker is identified in DICOM Software (treatment planning software, etc.), and a treatment plan is created using the skin marker as a reference point. However, since only skin markers are identified in medical images, the location of the electrode or electrode array that attaches the electrode to the actual target must be calculated in treatment planning simulation. In the case of skin markers, because they are 3D volumes, they are expressed as multiple points. Due to this problem, there is a problem in accuracy because the reference points for attachment are different, and (A) shows that in this medical image, skin markers are identified, but marks on the body are not. (B) shows that the skin marker cannot be expressed as a single point in a medical image. This example shows a medical image (Magnetic Resonant Imaging, MIR) containing skin markers, and shows that skin markers were identified in a total of 9 slides. Through this, the mark on the body is expressed as a single point, but the skin marker has a 3D volume and is expressed in multiple slices in the medical image.

Figure 3 shows that in detecting the second reference point for the purpose of this study, the center point of the skin marker cannot be found because the existing machine learning technology classifies the skin marker by tissue type (skin). (A) is the result of machine learning and shows that skin markers are also included in the external tissue of the human body when viewed in cross section. (B) shows that skin markers are included in the external tissue of the human body when looking at the above result as a 3D model.

Referring to FIG. 4, the present invention consists of (1) a segmentation stage and (2) a reference point detection stage (Fine Stage). (1) The initial segmentation step consists of a medical image data preprocessing step and a 3D volume of interest (VOI) automatic extraction step using a machine learning (convolutional neural network) deep learning model. (2) In the center point detection step, the volume of interest created above is divided into markers and tissue types, and the center point of the skin marker is used to create a new second reference point at the outermost edge of the tissue type. Here, the center point includes the skin marker and the center point when viewed from above the vertical plane of the outermost surface (External). There is a projection method as a method of creating a new second reference point, and it includes projection in three-dimensional space. At this time, the 3D volume medical image data can be one of CT, CBCT, MRI, and PET (Positron Emission Tomography), and any medical image data in which the patient's anatomical structure is obtained as 3D volume data is also possible. . The anatomical center point of the patient to be detected can be any center point included in the 3D volume medical image data, and includes a method of automatically detecting the center point by applying the same proposed method to all target center points without separate processing.

Figure 5 is a diagram illustrating the process (projection) of moving a point at the center point of a skin marker to a point on the skin, and the formula can be expressed as follows.

In vector notation, a plane can be expressed as a set of points such as p as follows.

n is the normal vector for the plane, and P0 is expressed as a point on the plane. (Notation A B represents the point product of vectors.

The vector equation for a line can be expressed as:

l is a vector in the direction of the line, I0 is a point on the line, and d is a real number. Substituting the equation for the line into the equation for the plane gives us

Expanding this,

If we summarize and express it for d, it is organized as follows.

Using the formula above, you can find the intersection point between the vector and the plane. Therefore, the second reference point can be obtained from the center point of the skin marker by applying the above formula.

Figure 6 shows the skin marker, the center point of the skin marker (The Feature Point of Skin Marker), and the detected second reference point (The Second Reference Point on the Skin) detected in the medical image using the method presented in the present invention. It's a picture.

Figure 7 is a diagram showing an example of using the detected second reference point. The center point of the electrode is based on the second reference point (The Second Reference Point on the Skin) detected by the above method in the electric field treatment planning software. The coordinates of Point of the electrode arrangement can be detected. Through this, multiple users who use skin markers to indicate the location of electrodes can use unified coordinates. As a result, by detecting the second reference point on the skin based on the center point of the skin marker identifiable in the medical image based on the displayed first reference point, the unity of the treatment plan can be improved, which increases the accuracy of the treatment plan and electrode attachment. . Therefore, not only professionals but also all medical personnel can use this system to share treatment plan results with each other.

Figure 8 is a diagram showing the data output stage to be output to the user.

(A) is a diagram showing the coordinates of the detected center point being output to the user. The center point information includes information indicating the location of the marker and coordinates (x, y, z) in three-dimensional space. Information output to the user may output the center point of each electrode based on the second reference point or the center point of the electrode array created with two or more electrodes.

(B) is a figure showing the output method of the detected segmented data. It can be output as Axial, Sagittal, Coronal, and 3D Volume Mesh, respectively. The output method is DICOM RTSS (Radiotherapy Structure Set) or NIFTI (Contour format). It can be output in (Neuroimaging informatics Technology Initiative) format.

Figure 9 shows two methods for applying machine learning to detect the second reference point. In the method of detecting the second reference point, the first method uses two machine learning methods and has two input data and output data in two steps. By dividing machine learning into two stages, users can edit data divided by skin marker and tissue type or input other data. The second method is machine learning that combines the above two stages of machine learning into one. It has one input data and one output data, and as a result, it shows that when one input data is input, a second reference point is finally detected. This method It can process a variety of data at high speed.

Claims (1)

Skin marker coordinate detection system.